U.S. patent application number 09/771963 was filed with the patent office on 2002-07-25 for system, method and article of manufacture for compiling and invoking c functions in hardware.
Invention is credited to Bowen, Matt.
Application Number | 20020100029 09/771963 |
Document ID | / |
Family ID | 25093472 |
Filed Date | 2002-07-25 |
United States Patent
Application |
20020100029 |
Kind Code |
A1 |
Bowen, Matt |
July 25, 2002 |
System, method and article of manufacture for compiling and
invoking C functions in hardware
Abstract
A method and computer program product are provided for compiling
a C function to a reconfigurable logic device. A function written
in a C programming language is received. The C function is compiled
into processor instructions, which are in turn used to generate
hardware configuration information. The hardware configuration
information is utilized to configure a Field Programmable Gate
Array (FPGA) for compiling the function to the FPGA. A system for
compiling a C function to a reconfigurable logic device is also
provided. The system includes receiving logic for receiving a
function written in a C programming language. Compiling logic is
used to compile the C function into processor instructions.
Conversion logic generates hardware configuration information from
the processor instructions. Configuring logic utilizes the hardware
configuration information to configure an FPGA such that the
function is compiled to the FPGA.
Inventors: |
Bowen, Matt; (Oxford,
GB) |
Correspondence
Address: |
C Douglas McDonald Esq
Carlton Fields Et AL
P O Box 3239
Tampa
FL
33601-3239
US
|
Family ID: |
25093472 |
Appl. No.: |
09/771963 |
Filed: |
January 29, 2001 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
09771963 |
Jan 29, 2001 |
|
|
|
09687011 |
Oct 12, 2000 |
|
|
|
60219754 |
Jul 20, 2000 |
|
|
|
Current U.S.
Class: |
717/140 ;
717/141 |
Current CPC
Class: |
G06F 8/41 20130101; G06F
8/447 20130101 |
Class at
Publication: |
717/140 ;
717/141 |
International
Class: |
G06F 009/45 |
Claims
What is claimed is:
1. A method for compiling a C function to a reconfigurable logic
device, comprising the steps of: (a) receiving a function written
in a C programming language; (b) compiling the C function into
processor instructions; (c) generating hardware configuration
information from the processor instructions; and (d) utilizing the
hardware configuration information for configuring a Field
Programmable Gate Array (FPGA) for compiling the function to the
FPGA.
2. A method as recited in claim 1, wherein the function in the FPGA
is shared amongst all its uses.
3. A method as recited in claim 1, wherein the configuration of the
FPGA is duplicated for each use.
4. A method as recited in claim 1, further comprising the step of
specifying N copies of the function for use M times.
5. A method as recited in claim 1, further comprising the step of
invoking the function utilizing a token.
6. A method as recited in claim 5, wherein the step of invoking the
function further includes the steps of passing the token to a start
signal, routing the start signal and call data to the function, and
storing the token in a wait sub-circuit until the function is
completed.
7. A computer program product for compiling a C function to a
reconfigurable logic device, comprising: (a) computer code for
receiving a function written in a C programming language; (b)
computer code for compiling the C function into processor
instructions; (c) computer code for generating hardware
configuration information from the processor instructions; and (d)
computer code for utilizing the hardware configuration information
for configuring a Field Programmable Gate Array (FPGA) for
compiling the function to the FPGA.
8. A computer program product as recited in claim 7, wherein the
function in the FPGA is shared amongst all its uses.
9. A computer program product as recited in claim 7, wherein the
configuration of the FPGA is duplicated for each use.
10. A computer program product as recited in claim 7, further
comprising computer code for specifying N copies of the function
for use M times.
11. A computer program product as recited in claim 7, further
comprising computer code for invoking the function utilizing a
token.
12. A computer program product as recited in claim 11, wherein the
computer code for invoking the function further includes computer
code for passing the token to a start signal, computer code for
routing the start signal and call data to the function, and
computer code for storing the token in a wait sub-circuit until the
function is completed.
13. A system for compiling a C function to a reconfigurable logic
device, comprising: (a) receiving logic for receiving a function
written in a C programming language; (b) compiling logic for
compiling the C function into processor instructions; (c)
conversion logic for generating hardware configuration information
from the processor instructions; (d) a Field Programmable Gate
Array (FPGA); and (e) configuring logic for utilizing the hardware
configuration information for configuring the FPGA for compiling
the function to the FPGA.
14. A system as recited in claim 13, wherein the function in the
FPGA is shared amongst all its uses.
15. A system as recited in claim 13, wherein the configuration of
the FPGA is duplicated for each use.
16. A system as recited in claim 13, further comprising logic for
specifying N copies of the function for use M times.
17. A system as recited in claim 13, further comprising control
logic for invoking the function utilizing a token.
18. A system as recited in claim 17, wherein the control logic for
invoking the function further includes logic for passing the token
to a start signal, logic for routing the start signal and call data
to the function, and logic for storing the token in a wait
sub-circuit until the function is completed.
Description
RELATED APPLICATIONS
[0001] This application is a continuation in part of U.S. patent
application entitled System, Method, and Article of Manufacture for
System Partitioning of a Reconfigurable Logic Device, Ser. No.
09/687011, filed Oct. 12, 2000, which claims priority from
Provisional U.S. Patent Application entitled System, Method, and
Article of Manufacture for System Partitioning of a Reconfigurable
Logic Device, serial No. 60/219754, filed Jul. 20, 2000, and which
are incorporated herein by reference for all purposes.
FIELD OF THE INVENTION
[0002] The present invention relates to a system for designing and
producing an electronic circuit having a desired functionality and
comprising both hardware which is dedicated to execution of certain
of the functionality and software-controlled machines for executing
the remainder of the functionality under the control of suitable
software.
BACKGROUND OF THE INVENTION
[0003] It is well known that software-controlled machines provide
great flexibility in that they can be adapted to many different
desired purposes by the use of suitable software. As well as being
used in the familiar general purpose computers, software-controlled
processors are now used in many products such as cars, telephones
and other domestic products, where they are known as embedded
systems.
[0004] However, for a given a function, a software-controlled
processor is usually slower than hardware dedicated to that
function. A way of overcoming this problem is to use a special
software-controlled processor such as a RISC processor which can be
made to function more quickly for limited purposes by having its
parameters (for instance size, instruction set etc.) tailored to
the desired functionality.
[0005] Where hardware is used, though, although it increases the
speed of operation, it lacks flexibility and, for instance,
although it may be suitable for the task for which it was designed
it may not be suitable for a modified version of that task which is
desired later. It is now possible to form the hardware on
reconfigurable logic circuits, such as Field Programmable Gate
Arrays (FPGA's) which are logic circuits which can be repeatedly
reconfigured in different ways. Thus they provide the speed
advantages of dedicated hardware, with some degree of flexibility
for later updating or multiple functionality.
[0006] In general, though, it can be seen that designers face a
problem in finding the right balance between speed and generality.
They can build versatile chips which will be software controlled
and thus perform many different functions relatively slowly, or
they can devise application-specific chips that do only a limited
set of tasks but do them much more quickly.
[0007] A compromise solution to these problems can be found in
systems which combine both dedicated hardware and also software.
The hardware is dedicated to particular functions, e.g. those
requiring speed, and the software can perform the remaining
functions. The design of such systems is known as hardware-software
codesign.
[0008] Within the design process, the designer must decide, for a
target system with a desired functionality, which functions are to
be performed in hardware and which in software. This is known as
partitioning the design. Although such systems can be highly
effective, the designer must be familiar with both software and
hardware design. It would be advantageous if such systems could be
designed by people who have familiarity only with software and
which could utilize the flexibility of configurable logic
resources.
SUMMARY OF THE INVENTION
[0009] In accordance with the invention, a method and computer
program product are provided for compiling a C function to a
reconfigurable logic device. A function written in a C programming
language is received. The C function is compiled into processor
instructions, which are in turn used to generate hardware
configuration information. The hardware configuration information
is utilized to configure a Field Programmable Gate Array (FPGA) for
compiling the function to the FPGA. Note that the methodology of
the present invention could also be applied to compile functions to
reconfigurable logic devices other than FPGAs. Handel-C is the
preferred programming language for carrying out the methodology of
the present invention and configuring the FPGA.
[0010] A system for compiling a C function to a reconfigurable
logic device is also provided. The system includes receiving logic
for receiving a function written in a C programming language.
Compiling logic is used to compile the C function into processor
instructions. Conversion logic generates hardware configuration
information from the processor instructions. Configuring logic
utilizes the hardware configuration information to configure an
FPGA such that the function is compiled to the FPGA.
[0011] In one embodiment of the present invention, the function is
a shared function. More particularly, the function in the FPGA is
shared amongst all its uses. In another embodiment of the present
invention, the configuration of the FPGA is duplicated for each
use, so that the function is used as an inline function. In yet
another embodiment of the present invention, the FPGA is configured
to provide an array of functions, where N copies of the function
are specified for use M times.
[0012] In a preferred embodiment of the present invention, a token
is used to invoke the function. Preferably, when invoking the
function, the token is passed to a start signal, the start signal
and call data are routed to the function, and the token is stored
in a wait sub-circuit until the function is completed.
BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The invention will be better understood when consideration
is given to the following detailed description thereof. Such
description makes reference to the annexed drawings wherein:
[0014] FIG. 1 is a flow diagram of a process for automatically
partitioning a behavioral description of an electronic system into
the optimal configuration of hardware and software according to a
preferred embodiment of the present invention;
[0015] FIG. 2 is a flow diagram schematically showing the codesign
system of one embodiment of the invention;
[0016] FIG. 3 illustrates the compiler objects which can be defined
in one embodiment of the invention;
[0017] FIG. 4 is a block diagram of the platform used to implement
the second example circuit produced by an embodiment of the
invention;
[0018] FIG. 5 is a picture of the circuit of FIG. 4;
[0019] FIG. 6 is a block diagram of the system of FIG. 4;
[0020] FIG. 7 is a simulation of the display produced by the
example of FIGS. 4 to 6;
[0021] FIG. 8 is a block diagram of a third example target
system;
[0022] FIGS. 9A-D are a block diagram showing a dependency graph
for calculation of the variables in the FIG. 8 example;
[0023] FIG. 10 is a schematic diagram of a hardware implementation
of one embodiment of the present invention;
[0024] FIG. 11 is a flow diagram of a process for compiling a C
function to a reconfigurable logic device;
[0025] FIG. 12 is a diagram of a function call sub-circuit
according to an embodiment of the present invention; and
[0026] FIG. 13 is an illustration of a pass by value sub-circuit
according to one embodiment of the present invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0027] The present invention provides a hardware/software codesign
system which can target a system in which the hardware or the
processors to run the software can be customized according to the
functions partitioned to it. Thus rather than the processor or
hardware being fixed (which effectively decides the partitioning),
the codesign system of this invention includes a partitioning means
which flexibly decides the partitioning while varying the
parameters of the hardware or processor to obtain both an optimal
partitioning and optimal size of hardware and processor.
[0028] In more detail it provides a codesign system for producing a
target system having resources to provide specified functionality
by:
[0029] (a) operation of dedicated hardware; and
[0030] (b) complementary execution of software on
software-controlled machines;
[0031] The codesign system comprising means for receiving a
specification of the functionality, partitioning means for
partitioning implementation of the functionality between (a) and
(b) and for customizing the hardware and/or the machine in
accordance with the selected partitioning of the functionality.
[0032] Thus the target system is a hybrid hardware/software system.
It can be formed using configurable logic resources in which case
either the hardware or the processor, or both, can be formed on the
configurable logic resources (e.g. an FPGA).
[0033] In one embodiment of the invention the partitioning means
uses a genetic algorithm to optimize the partitioning and the
parameters of the hardware and the processor. Thus, it generates a
plurality of different partitions of the functionality of the
target system (varying the size of the hardware and/or the
processor between the different partitions) and estimates the speed
and size of the resulting system. It then selects the optimal
partitioning on the basis of the estimates. In the use of a genetic
algorithm, a variety of partitions are randomly generated, the poor
ones are rejected, and the remaining ones are modified by combining
aspects of them with each other to produce different partitions.
The speed and size of these are then assessed and the process can
be repeated until an optimal partition is produced.
[0034] The invention is applicable to target systems which use
either customizable hardware and a customizable processor, or a
fixed processor and customizable hardware, or fixed hardware and a
customizable processor. Thus the customizable part could be formed
on an FPGA, or, for instance, an ASIC. The system may include
estimators for estimating the speed and size of the hardware and
the software controlled machine and may also include an interface
generator for generating interfaces between the hardware and
software. In that case the system may also include an estimator for
estimating the size of the interface. The partitioning means calls
the estimators when deciding on an optimum partitioning.
[0035] The software-controlled machine can comprise a CPU and the
codesign system comprises means for generating a compiler for the
CPU as well as means for describing the CPU where it is to be
formed on customizable logic circuits.
[0036] The codesign system can further comprise a hardware compiler
for producing from those parts of the specification partitioned to
hardware a register transfer level description for configuring
configurable logic resources (such as an FPGA). It can further
include a synthesizer for converting the register transfer level
description into a net list.
[0037] The system can include a width adjuster for setting and
using a desired data word size, and this can be done at several
points in the desired process as necessary.
[0038] Another aspect of the invention provides a hardware/software
codesign system which receives a specification of a target system
in the form of behavioral description, i.e. a description in a
programming language such as can be written by a computer
programmer, and partitions it and compiles it to produce hardware
and software.
[0039] The partitioning means can include a parser for parsing the
input behavioral description. The description can be in a familiar
computer language such as C, supplemented by a plurality of
predefined attributes to describe, for instance, parallel execution
of processes, an obligatory partition to software or an obligatory
partition to hardware. The system is preferably adapted to receive
a declaration of the properties of at least one of the hardware and
the software-controlled machine, preferably in an object-oriented
paradigm. It can also be adapted such that some parts of the
description can be at the register transfer level, to allow closer
control by the user of the final performance of the target
system.
[0040] Thus, in summary, the invention provides a hardware/software
codesign system for making an electronic circuit which includes
both dedicated hardware and software controlled resources. The
codesign system receives a behavioral description of the target
electronic system and automatically partitions the required
functionality between hardware and software, while being able to
vary the parameters (e.g. size or power) of the hardware and/or
software. Thus, for instance, the hardware and the processor for
the software can be formed on an FPGA, each being no bigger than is
necessary to form the desired functions. The codesign system
outputs a description of the required processor (which can be in
the form of a net list for placement on the FPGA), machine code to
run on the processor, and a net list or register transfer level
description of the necessary hardware. It is possible for the user
to write some parts of the description of the target system at
register transfer level to give closer control over the operation
of the target system, and the user can specify the processor or
processors to be used, and can change, for instance, the
partitioner, compilers or speed estimators used in the codesign
system. The automatic partitioning can be performed by using an
optimization algorithm, e.g. a genetic algorithm, which generates a
partitioning based on estimates of performance.
[0041] The invention also allows the manual partition of systems
across a number of hardware and software resources from a single
behavioral description of the system. This provision for manual
partitioning, as well as automatic partitioning, gives the system
great flexibility.
[0042] The hardware resources may be a block that can implement
random hardware, such as an FPGA or ASIC; a fixed processor, such
as a microcontroller, DSP, processor, or processor core; or a
customizable processor which is to be implemented on one of the
hardware resources, such as an FPGA-based processor. The system
description can be augmented with register transfer level
descriptions, and parameterized instantiations of both hardware and
software library components written in other languages.
[0043] The sort of target systems which can be produced
include:
[0044] a fixed processor or processor core, coupled with custom
hardware;
[0045] a set of customizable (e.g. FPGA-based) processors and
custom hardware;
[0046] a system on a chip containing fixed processors and an FPGA;
and
[0047] a PC containing an FPGA accelerator board.
[0048] The use of the advanced estimation techniques in specific
embodiments of the invention allows the system to take into account
the area of the processor that will be produced, allowing the
targeting of customizable processors with additional and removable
instructions, for example. The estimators also take into account
the speed degradation produced when the logic that a fixed hardware
resource must implement nears the resource's size limit. This is
done by the estimator reducing the estimated speed as that limit is
reached. Further, the estimators can operate on both the design
before partitioning, and after partitioning. Thus high level
simulation, as well as simulation and estimation after
partitioning, can be performed.
[0049] Where the system is based on object oriented design, this
allows the user to add new processors quickly and to easily define
their compilers.
[0050] The part of the system which compiles the software can
transparently support additional or absent instructions for the
processor and so is compatible with the parameterization of the
processor.
[0051] Preferably, the input language supports variables with
arbitrary widths, which are then unified to a fixed width using a
promotion scheme, and then mapped to the widths available on the
target system architecture.
[0052] Further in one embodiment of the invention it is possible
for the input description to include both behavioral and register
transfer level descriptions, which can both be compiled to
software. This gives support for very fast simulation and allows
the user control of the behavior of the hardware on each clock
cycle.
[0053] FIG. 1 is a flow diagram of a process 100 for automatically
partitioning a behavioral description of an electronic system into
the optimal configuration of hardware and software according to a
preferred embodiment of the present invention. In operation 102,
the system receives a behavioral description of the electronic
system and, in operation 104, determines the optimal required
functionality between hardware and software. In operation 106, that
functionality is partitioned preferably while varying the
parameters (e.g. size or power) of the hardware and/or software.
Thus, for instance, the hardware and the processors for the
software can be formed on a reconfigurable logic device, each being
no bigger than is necessary to form the desired functions.
[0054] The codesign system outputs a description of the required
processors, machine code to run on the processors, and a net list
or register transfer level description of the necessary hardware.
It is possible for the user to write some parts of the description
of the system at register transfer level to give closer control
over the operation of the system, and the user can specify the
processor or processors to be used, and can change, for instance,
the partitioner, compilers or speed estimators used in the codesign
system. The automatic partitioning is formed by using a genetic
algorithm which estimates the performance of randomly generated
different partitions and selects an optimal one of them.
[0055] This description will later refer to specific examples of
the input behavioral or register transfer level description of
examples of target systems. These examples are reproduced in
Appendices, namely:
[0056] Appendix 1 is an exemplary register transfer level
description of a simple processor.
[0057] Appendix 2 is a register transfer level description of the
main process flow in the example of FIGS. 4 to 6.
[0058] Appendix 3 is the input specification for the target system
of FIG. 8.
[0059] The flow of the codesign process in an embodiment of the
invention is shown in FIG. 2 and will be described below. The
target architecture for this system is an FPGA containing one or
more processors, and custom hardware. The processors may be of
different architectures, and may communicate with each other and
with the custom hardware.
[0060] The Input Language
[0061] In this embodiment the user writes a description 202 of the
system in a C-like language, which is actually ANSI C with some
additions which allow efficient translation to hardware and
parallel processes. This input description will be compiled by the
system 200 of FIG. 2. The additions to the ANSI C language include
the following:
[0062] Variables are declared with explicit bit widths and the
operators working on the variables work with an arbitrary
precision. This allows efficient implementation in hardware. For
instance a statement which declares the width of variables (in this
case the program counter pc, the instruction register ir, and the
top of stack tos) is as follows:
[0063] unsigned 12 pc, ir, tos
[0064] The width of the data path of the processor in the target
system may be declared, or else is calculated by the partitioner
208 as the width of the widest variable which it uses.
[0065] The "par" statement has been added to describe process-level
parallelism. The system can automatically extract fine-grained
parallelism from the C-like description but generating
coarse-grained parallelism automatically is far more difficult.
Consequently the invention provides this attribute to allow the
user to express parallelism in the input language using the "par"
statement which specifies that a following list of statements is to
be executed in parallel. For example, the expression:
1 Par { parallel_port(port); SyncGeno; }
[0066] means that two sub-routines, the first of which is a driver
for a parallel port and the second of which is a sync generator for
a video display are to be executed in parallel. All parts of the
system will react to this appropriately.
[0067] Channels can be declared and are used for blocking,
point-to-point synchronized communication as used in occam (see G.
Jones. Programming in occam. Prentice Hall International Series in
Computer Science, 1987, which is hereby incorporated by reference)
with a syntax like a C function call. The parallel processes can
use the channels to perform distributed assignment. Thus parallel
processes can communicate using blocking channel communication. The
keyword "chan" I declares these channels. For example,
[0068] chan hwswchan; i I
[0069] declares a channel along which variables will be sent and
received between the hardware and software parts of the system.
Further,
[0070] send (channel 1, a)
[0071] is a statement which sends the value of variable a down
channel 1; and receive (channel 2, b) is a statement which assigns
the value received along channel 2 to variable b.
[0072] The hardware resources available are declared. The resources
may be a customizable processor, a fixed processor, or custom
hardware. The custom hardware may be of a specific architecture,
such as a Xilinx FPGA. Further, the architecture of the target
system can be described in terms of the available functional units
and their interconnection.
[0073] To define the architecture "platforms" and "channels" are
defined. A platform can be hard or soft. A hard platform is
something that is fixed such as a Pentium processor or an FPGA. A
soft platform is something that can be configured like an
FPGA-based processor. The partitioner 208 understands the keywords
"hard" and "soft", which are used for declaring these platforms and
the code can be implemented on any of these.
[0074] This particular embodiment supports the following hard
platforms:
[0075] Xilinx 4000 series FPGAs (e.g. the Xilinx 4085 below);
[0076] Xilinx Virtex series FPGAs;
[0077] Altera Flex and APEX PLDs;
[0078] Processor architectures supported by ANSI C compilers;
[0079] and the following soft platforms each of which is associated
with one of the parameterizable processors mentioned later:
[0080] FPGAStackProc, FPGAParallelStackProc, FPGAMips.
[0081] An attribute can be attached to a platform when it is
declared:
[0082] platform (PLATFORMS ) y t c
[0083] For a hard platform the attribute PLATFORMS contains one
element: the architecture of the hard platform. In this embodiment
this may be the name of a Xilinx 3000 or 4000 series FPGA, an
Altera FPGA, or an x86 processor.
[0084] For a soft platform, PLATFORMS is a pair. The first element
is the architecture of the platform:
[0085] FPGAStackProc, FPGAParallelStackProc or FPGAMips
[0086] and the second is the name of the previously declared
platform on which the new platform is implemented.
[0087] Channels can be declared with an implementation, and as only
being able to link previously declared platforms. The system 200
recognizes the following channel implementations:
[0088] PCIBus--a channel implemented over a PCI bus between an FPGA
card and a PC host.
[0089] FPGAChan--a channel implemented using wires on the FPGA.
[0090] The following are the attributes which can be attached to a
channel when it is declared:
[0091] type (CHANNELTYPE)
[0092] This declares the implementation of the channel. Currently
CHANNELTYPE may be PCIBus or FPGAChan. FPGAChan is the default.
[0093] from(PLATFORM)
[0094] PLATFORM is the name of the platform which can send down the
channel.
[0095] to (PLATFORM)
[0096] PLATFORM is the name of the platform which can receive from
the channel.
[0097] The system 200 checks that the declared channels and the
platforms that use them are compatible. The communication
mechanisms which a given type of channel can implement are built
into the system. New mechanisms can be added by the user, in a
similar way to adding new processors as will be explained
below.
[0098] Now an example of an architecture will be given.
[0099] Example Architecture
2 /* Architectural Declarations */ // the 4085 is a hard platform
-- call this one meetea board hard meeteaBoard
-attribute_((platform(Xilinx4085))); // the pentium is a hard
platform -- call this one hostProcessor hard hostProcessor
attribute- ((platform(Pentium))); // proci is a soft platform which
is implemented // on the FPGA on the meetea board soft proci
attribute- ((platform(FpgaStackProc, meeteaBoard)));
[0100] Example Program
3 void main() { // channel1 is implemented on a PCIBus I // and can
send data from hostProcessor to meetea board chan channel1
attribute- ((type(PCIBus), from(hostProcessor), to(meeteaBoard)));
// channel2 is implemented on the FPGA chan channel2,attribute-
((type(FPGAChan))); /* the code */ par { // code which can be
assigned to // either hostProcessor (software), // or prod
(software of reconfigurable processor), // or meetea board
(hardware), // or left unassigned (compiler decides). //
Connections between hostProcessor // and prod or meetea must be
over the PCI Bus // (channel1) // Connections between procl and
hardware // must be over the FPGA channel (channel2)
[0101] Attributes are also added to the input code to enable the
user to specify whether a block is to be put in hardware or
software and for software the attribute also specifies the target
processor. The attribute is the name of the target platform. For
example:
4 { int a, b; a = a + b; } attribute-
((platform(hostProcessor)))
[0102] assigns the operation a+b to Host Processor.
[0103] For hardware the attribute also specifies whether the
description is to be interpreted as a register transfer (RT) or
behavioral level description. The default is behavioral. For
example:
5 { int a, b; par { b = a + b; a b, } }
,attribute-((platform(meeteaBoard),level(R- TL)))
[0104] would be compiled to hardware using the RTL compiler, which
would guarantee that the two assignments happened on the same clock
cycle.
[0105] Thus parts of the description which are to be allocated to
hardware can be written by the user at a register transfer level,
by using a version of the input language with a well defined timing
semantics (for example Handel-C or another RTL language), or the
scheduling decisions (i.e. which operations happen on which clock
cycle) can be left to the compiler. Thus using these attributes a
block of code may be specifically assigned by the user to one of
the available resources. Soft resources may themselves be assigned
to hardware resources such as an FPGA-based processor. The
following are the attributes which can be attached to a block of
code:
[0106] platform(PLATFORM)
[0107] PLATFORM is the name of the platform on which the code will
be implemented. This implies the compiler which will be used to
compile that code.
[0108] level(LEVEL)
[0109] LEVEL is Behavioral or RTL. Behavioral descriptions will be
scheduled and may be partitioned. RTL descriptions are passed
straight through to the RTL synthesizer e.g. a Handel-C
compiler.
[0110] cycles(NUMBER)
[0111] NUMBER is a positive integer. Behavioral descriptions will
be scheduled in such a way that the block of code will execute
within that number of cycles, when possible. An error is generated
if it is not possible.
[0112] Thus the use of this input language which is based on a
known computer language, in this case C, but with the additions
above allows the user, who could be a system programmer, to write a
specification for the system in familiar behavioral terms like a
computer program. The user only needs to learn the additions above,
such as how to declare parallelism and to declare the available
resources to be able to write the input description of the target
system.
[0113] This input language is input to the parser 204 which parses
and type checks the input code, and performs some syntax level
optimizations, (in a standard way for parsers), and attaches a
specific compiler to each block of code based on the attributes
above. The parser 204 uses standard techniques [Aho, Sethi and
Ullman; "Compilers Principles, Techniques, and Tools"; Addison
Wesley known as "The Dragon Book", which is hereby incorporated by
reference] to turn the system description in the input language
into an internal data structure, the abstract syntax tree which can
be supplied to the partitioner 208.
[0114] The width adjuster 206 uses C-techniques to promote
automatically the arguments of operators to wider widths such that
they are all of the same width for instance by concatenating them
with zeros. Thus this is an extension of the promotion scheme of
the C language, but uses arbitrary numbers of bits. Further
adjustment is carried out later in the flow at 206a and 206b, for
instance by ANDing them with a bit mask. Each resource has a list
of widths that it can support. For example a 32 bit processor may
be able to carry out 8, 16 and 32 bit operations. Hardware may be
able to support any width, or a fixed width datapath operator may
have been instantiated from a library. The later width adjustment
modules 206a and 206b insert commands to enable the width of
operation in the description to be implemented correctly using the
resources available.
[0115] Hardware/Software Partitioning
[0116] The partitioner 208 generates a control/data-flow graph
(CDFG) from the abstract syntax tree, for instance using the
techniques described in G. de Michelli "Synthesis and Optimization
of Digital Circuits"; McGraw-Hill, 1994 which is hereby
incorporated by reference. It then operates on the parts of the
description which have not already been assigned to resources by
the user. It groups parts of the description together into blocks,
"partitioning blocks", which are indivisible by the partitioner.
The size of these blocks is set by the user, and can be any size
between a single operator, and a top-level process. Small blocks
tend to lead to a slow more optimal partition; large blocks tend to
lead to a faster less optimal partition.
[0117] The algorithm used in this embodiment is described below but
the system is designed so that new partitioning algorithms can
easily be added, and the user can choose which of these
partitioning algorithms to use. The algorithms all assign each
partitioning block to one of the hardware resources which has been
declared.
[0118] The algorithms do this assignment so that the total
estimated hardware area is smaller than the hardware resources
available, and so that the estimated speed of the system is
maximized.
[0119] The algorithm implemented in this embodiment of the system
is a genetic algorithm for instance as explained in D. E. Goldberg,
"Genetic Algorithms in Search, Optimization and Machine learning",
Addison-Wesley, 1989 which is hereby incorporated by reference. The
resource on which each partitioning block is to be placed
represents a gene and the fitness function returns infinity for a
partitioning which the estimators say will not fit in the available
hardware; otherwise it returns the estimated system speed.
Different partitions are generated and estimated speed found. The
user may set the termination condition to one of the following:
[0120] 1) when the estimated system speed meets a given
constraint;
[0121] 2) when the result converges, i.e. the algorithm has not
resulted in improvement after a user-specified number of
iterations;
[0122] 3) when the user terminates the optimization manually.
[0123] The partitioner 208 uses estimators 220, 222, and 224 to
estimate the size and speed of the hardware, software and
interfaces as described below.
[0124] It should be noted from FIG. 2 that the estimators and the
simulation and profiling module 220 can accept a system description
from any level in the flow. Thus it is possible for the input
description, which may include behavioral and register transfer
level parts, to be compiled to software for simulation and
estimation at this stage. Further, the simulator can be used to
collect profiling information for sets of typical input data, which
will be used by the partitioner 208 to estimate data dependent
values, by inserting data gathering operations into the output
code.
[0125] Hardware Estimation
[0126] The estimator 222 is called by the partitioner 208 for a
quick estimation of the size and speed of the hardware parts of the
system using each partition being considered. Data dependent values
are estimated using the average of the values for the sets of
typical input data supplied by the user.
[0127] To estimate the speed of hardware, the description is
scheduled using a call to the behavioral synthesizer 212. The user
can choose which estimation algorithm to use, which gives a choice
between slow accurate estimation and faster less accurate
estimation. The speed and area of the resulting RTL level
description is then estimated using standard techniques. For FPGAs
the estimate of the speed is then decreased by a non-linear factor
determined from the available free area, to take into account the
slower speed of FPGA designs when the FPGA is nearly full.
[0128] Software Estimation
[0129] If the software is to be implemented on a fixed processor,
then its speed is estimated using the techniques described in J.
Madsen and J. Grode and P. V. Knudsen and M. E. Petersen and A.
I-Iaxthausen, "LYCOS: the Lyngby Co-Synthesis System, Design
Automation of Embedded Systems, 1977, volume 2, number 2, (Madsen
et al) which is hereby incorporated by reference. The area of
software to be implemented on a fixed processor is zero.
[0130] If the target is customizable processors to be compiled by
the system itself then a more accurate estimation of the software
speed is used which models the optimizations that the software
compiler 216 uses. The area and cycle time of the processor is
modeled using a function which is written for each processor, and
expresses the required values in terms of the values of the
processor's parameterizations, such as the set of instructions that
will be used, the data path and instruction register width and the
cache size.
[0131] Interface Synthesis and Estimation
[0132] Interfaces between the hardware and software are
instantiated by the interface cosynthesizer 210 from a standard
library of available communication mechanisms. Each communication
mechanism is associated with an estimation function, which is used
by the partitioner to cost the software and hardware speed and area
required for given communication, or set of communications.
Interfaces which are to be implemented using a resource which can
be parameterized (such as a channel on an FPGA), are synthesized
using the parameterizations decided by the partitioner. For
example, if a transfer of ten thousand 32 bit values over a PCI bus
was required, a DMA transfer from the host to an FPGA card's local
memory might be used.
[0133] Compilation
[0134] The compiler parts of the system are designed in an object
oriented way, and actually provide a class hierarchy of compilers,
as shown in FIG. 3. Each node in the tree shows a class which is a
subclass of its parent node. The top-level compiler class 302
provides methods common to both the hardware and software flows,
such as the type checking, and a system-level simulator used for
compiling and simulating the high-level description. These methods
are inherited by the hardware and software compilers 304, 306, and
may be used or overridden. The compiler class also specifies other,
virtual, functions which must be supplied by its subclasses. So the
compile method on the hardware compiler class compiles the
description to hardware by converting the input description to an
RTL description; the compile method on the Processor A compiler
compiles a description to machine code which can run on Processor
A.
[0135] There are two ways in which a specific compiler can be
attached to a specific block of code:
[0136] A) In command line mode. The compiler is called from the
command line by the attributes mentioned above specifying which
compiler to use for a block of code.
[0137] B) Interactively. An interactive environment is provided,
where the user has access to a set of functions which the user can
call, e.g. to estimate speed and size of hardware and software
implementations, manually attach a compiler to a block of code, and
call the simulator. This interactive environment also allows
complex scripts, functions and macros to be written and saved by
the user for instance so that the user can add a new partitioning
algorithm.
[0138] The main compilation stages of the process flow are software
or hardware specific. Basically at module 212 the system schedules
and allocates any behavioral parts of the hardware description, and
at module 216 compiles the software description to assembly code.
At module 218 it also writes a parameterized description of the
processors to be used, which may also have been designed by the
user. These individual steps will be explained in more detail.
[0139] Hardware Compilation
[0140] The parts of the description to be compiled into hardware
use a behavioral synthesis compiler 212 using the techniques of De
Michelli mentioned above. The description is translated to a
control/data flow graph, scheduled (i.e. what happens on each clock
cycle is established) and bound (i.e. which resources are used for
which operations is established), optimized, and then an RT-level
description is produced.
[0141] Many designers want to have more control over the timing
characteristics of their hardware implementation. Consequently the
invention also allows the designer to write parts of the input
description corresponding to certain hardware at the register
transfer level, and so define the cycle-by-cycle behavior of that
hardware.
[0142] This is done by using a known RT-level description with a
well-defined timing semantics such as Handel-C. In such a
description each assignment takes one clock cycle to execute,
control structures add only combinational delay, and communications
take one clock cycle as soon as both processes are ready. With the
invention an extra statement is added to this RT-level version of
the language: "delay" is a statement which uses one clock cycle but
has no other effect. Further, the "par" attribute may again be used
to specify statements which should be executed in parallel.
[0143] Writing the description at this level, together with the
ability to define constraints for the longest combinational path in
the circuit, gives the designer close control of the timing
characteristics of the circuit when this is necessary. It allows,
for example, closer reasoning about the correctness of programs
where parallel processes write to the same variable. This extra
control has a price: the program must be refined from the more
general C description, and the programmer is responsible for
thinking about what the program is doing on a cycle-by-cycle basis.
An example of a description of a processor at this level will be
discussed later.
[0144] The result of the hardware compilation by the behavioral
synthesizer 212 is an RTL description which can be output to a RTL
synthesis system 214 using a hardware description language (e.g.
Handel-C or VHDL), or else synthesized to a gate level description
using the techniques of De Michelli.
[0145] RTL synthesis optimizes the hardware description, and maps
it to a given technology. This is performed using standard
techniques.
[0146] Software Compilation
[0147] The software compiler 216 largely uses standard techniques
[e.g. from Aho, Sethi and Ullman mentioned above]. In addition,
parallelism is supported by mapping the invention's CSP-like model
of parallelism and communication primitives into the target model.
For instance channels can mapped to blocks of shared memory
protected by semaphores. CSP is described in C. A. R. Hoare
"Communicating sequential processes." Prentice-Hall International
series in computing science. Prentice-Hall International, Englewood
Cliffs, N.J. which is hereby incorporated by reference.
[0148] Compound operations which are not supported directly by the
processor are decomposed into their constituent parts, or mapped to
operations on libraries. For example multiply can be decomposed
into shifts and adds. Greedy pattern matching is then used to map
simple operations into any more complex instructions which are
supported by the processor. Software can also be compiled to
standard ANSI C, which can then be compiled using a standard
compiler. Parallelism is supported by mapping the model in the
input language to the model of parallelism supported by the C
compiler, libraries and operating system being used.
[0149] The software compiler is organized in an object oriented way
to allow users to add support for different processors (see FIG. 3)
and for processor parameterizations. For example, in the processor
parameterize 218 unused instructions from the processor description
are automatically removed, and support for additional instructions
can be added. This embodiment of the invention, includes some
prewritten processor descriptions which can be selected by the
user. It contains parameterized descriptions of three processors,
and the software architecture is designed so that it is easy for
developers to add new descriptions which can be completely new or
refinements of these. The three processors provided are
[0150] A Mips-like processor, similar to that described in
[Patterson and Hennessy, Computer Organization and Design, 2"d
Edition, Morgan Kauffman].
[0151] A 2-cycle non-pipelined stack-based processor (see
below).
[0152] A more sophisticated multicycle non-pipelined stack-based
processor, with a variable number of cycles per instruction, and
hardware support for parallelism and channels.
[0153] Thus the software compiler supports many processor
parameterizations. More complex and unexpected modifications are
supported by virtue of the object oriented design of the compiler,
which allows small additions to be made easily by the user. Most of
the mapping functions can be inherited from existing processor
objects, minor additions can be made a function used to calculate
the speed and area of processor given the parameterizations of the
processor and a given program.
[0154] The output of the software compilation/processor
parameterization process is machine code to run on the processor
together with a description of the processor to be used (if it is
not a standard one).
[0155] Co-simulation and Estimation
[0156] The scheduled hardware, register transfer level hardware,
software and processor descriptions are then combined. This allows
a cycle-accurate co-simulation to be carried out, e.g. using the
known Handel-C simulator, though a standard VHDL or Verilog
simulator and compiler could be used.
[0157] Handel-C provides estimation of the speed and area of the
design, which is written as an HTML file to be viewed using a
standard browser, such as Netscape. The file shows two versions of
the program: in one each statement is colored according to how much
area it occupies, and in the other according to how much
combinational delay it generates. The brighter the color for each
statement, the greater the area or delay. This provides a quick
visual feedback to the user of the consequences of design
decisions.
[0158] The Handel-C simulator is a fast cycle-accurate simulator
which uses the C-like nature of the specification to produce an
executable which simulates the design. It has an X-windows
interface which allows the user to view VGA video output at about
one frame per second.
[0159] When the user is happy with the RT-level simulation and the
design estimates then the design can be compiled to a netlist. This
is then mapped, placed and routed using the FPGA vendor's
tools.
[0160] The simulator can be used to collect profiling information
for sets of typical input data, which will be used by the
partitioner 208 to estimate data dependent values, by inserting
data gathering operations into the output code.
[0161] Implementation Language
[0162] The above embodiment of the system was written in objective
CAML which is a strongly typed functional programming language
which is a version of ML but obviously it could be written in other
languages such as C.
[0163] Provable Correctness
[0164] A subset of the above system could be used to provide a
provably correct compilation strategy. This subset would include
the channel communication and parallelism of OCCAM and CSP. A
formal semantics of the language could be used together with a set
of transformations and a mathematician, to develop a provably
correct partitioning and compilation route.
[0165] Some examples of target systems designed using the invention
will now be described.
EXAMPLE 1
Processor Design
[0166] The description of the processor to be used to run the
software part of the target system may itself be written in the
C-like input language and compiled using the codesign system. As it
is such an important element of the final design most users will
want to write it at the register transfer level, in order to
hand-craft important parts of the design. Alternatively the user
may use the predefined processors, provided by the codesign system
or write the description in VHDL or even at gate level, and merge
it into the design using an FPGA vendor's tools.
[0167] With this system the user can parameterize the processor
design in nearly any way that he or she wishes as discussed above
in connection with the software compilation and as detailed
below.
[0168] The first processor parameterization to consider is removing
redundant logic. Unused instructions can be removed, along with
unused resources, such as the floating point unit or expression
stack.
[0169] The second parameterization is to add resources. Extra RAMS
and ROMs can be added. The instruction set can be extended from
user assigned instruction definitions. Power-on bootstrap
facilities can be added.
[0170] The third parameterization is to tune the size of the used
resources. The bit widths of the program counter, stack pointer,
general registers and the opcode and operand portions of the
instruction register can be set. The size of internal memory and of
the stack or stacks can be set, the number and priorities of
interrupts can be defined, and channels needed to communicate with
external resources can be added. This freedom to add communication
channels is a great benefit of codesign using a parametrizable
processor, as the bandwidth between hardware and software can be
changed to suit the application and hardware/software
partitioning.
[0171] Finally, the assignment of opcodes can be made, and
instruction decoding rearranged.
[0172] The user may think of other parameterizations, and the
object oriented processor description allows this. The description
of a very simple stack-based processor in this style (which is
actually one of the pre-written processors provided by the codesign
system for use by the user) is listed in Appendix 1.
[0173] Referring to Appendix 1, the processor starts with a
definition of the instruction width, and the width of the internal
memory and stack addresses. This is followed by an assignment of
the processor opcodes. Next the registers are defined; the
declaration "unsigned x y, z" declares unsigned integers y and z of
width x. The program counter, instruction register and top-of-stack
are the instruction width; the stack pointer is the width of the
stack.
[0174] After these declarations the processor is defined. This is a
simple non-pipelined two-cycle processor. On the first cycle (the
first three-line "par"), the next instruction is fetched from
memory, the program counter is incremented, and the top of the
stack is saved. On the second cycle the instruction is decoded and
executed. In this simple example a big switch statement selects the
fragment of code which is to be executed.
[0175] This simple example illustrates a number of points. Various
parameters, such as the width of registers and the depth of the
stack can be set. Instructions can be added by including extra
cases in the switch statement. Unused instructions and resources
can be deleted, and opcodes can be assigned.
[0176] The example also introduces a few other features of the
register transfer level 30 language such as rom and ram
declarations.
EXAMPLE 2
Video Game
[0177] To illustrate the use of the invention using an application
which is small enough to describe easily a simple Internet video
game was designed. The target system is a video game in which the
user can fly a plane over a detailed background picture. Another
user can be dialed up, and the screen shows both the local plane
and a plane controlled remotely by the other user. The main
challenge for the design is that the system must be implemented on
a single medium-sized FPGA.
[0178] Implementation Platform
[0179] The platform for this application was a generic and simple
FPGA-based board. A block diagram of the board 400, a Hammond
board, is shown in FIG. 4, and a graphical depiction of the board
400 is shown in FIG. 5.
[0180] The Hammond board contains a Xilinx 4000 series FPGA and 256
kb synchronous static RAM. Three buttons provide a simple input
device to control the plane; alternatively a standard computer
keyboard can be plugged into the board. There is a parallel port
which is used to configure the FPGA, and a serial port. The board
can be clocked at 20 MHz from a crystal, or from a PLL controlled
by the FPGA. Three groups of four pins of the FPGA are connected to
a resistor network which gives a simple digital to analogue
converter, which can be used to provide 12 bit VGA video by
implementing a suitable sync generator on the FPGA. Problem
description and discussion The specification of the video game
system is as follows:
[0181] The system must dial up an Internet service provider, and
establish a connection with the remote game. which will be running
on a workstation.
[0182] The system must display a reconfigurable background
picture.
[0183] The system must display on a VGA monitor a picture of two
planes: the local plane and the remote plane. The position of the
local plane will be controlled by the buttons on the Hammond
board.
[0184] The position of the remote plane will be received over the
dialup connection every time it changes.
[0185] The position of the local plane will be sent over the
dialup, connection every time it changes.
[0186] This simple problem combines some hard timing constraints,
such as sending a stream of video to the monitor, with some complex
tasks without timing constraints, such as connecting to the
Internet service provider. There is also an illustration of
contention for a shared resource, which will be discussed
later.
[0187] System Design
[0188] A block diagram of the system 600 is shown in FIG. 6. The
system design decisions were quite straightforward. A VGA monitor
602 is plugged straight into the Hammond board 400. To avoid the
need to make an electrical connection to the telephone network a
modem 604 can be used, and plugged into the serial port of the
Hammond board. Otherwise it is quite feasible to build a simple
modem in the FPGA.
[0189] The subsystems required are:
[0190] serial port interface,
[0191] dial up,
[0192] establishing the network connection,
[0193] sending the position of the local plane,
[0194] receiving the position of the remote plane,
[0195] displaying the background picture,
[0196] displaying the planes.
[0197] A simple way of generating the video is to build a sync
generator in the FPGA, and calculate and output each pixel of VGA
video at the pixel rate. The background picture can be stored in a
"picture RAM". The planes can be stored. As a set of 8.times.8
characters in a "character generator ROM", and the contents of each
of the characters' positions on the screen stored in a "character
location RAM.
[0198] Hardware/Software Partitioning
[0199] The hardware portions of the design are dictated by the need
of some part of the system to meet tight timing constraints. These
are the video generation circuitry and the port drivers.
Consequently these were allocated to hardware, and their C
descriptions written at register transfer level to enable them to
meet the timing constraints. The picture RAM and the character
generator ROM and character location RAM were all stored in the
Hammond board RAM bank as the size estimators showed that there
would be insufficient space on the FPGA.
[0200] The parts of the design to be implemented in software are
the dial-up and negotiation, establishing the network, and
communicating the plane locations. These are non-time critical, and
so can be mapped to software. The program is stored in the RAM
bank, as there is not space for the application code in the FPGA.
The main function is shown in Appendix 2. The first two lines
declare some communication channels. Then the driver for the
parallel port and sync generator are started, and the RAM is
initialized with the background picture, the character memory and
the program memory. The parallel communicating hardware and
software process are then started, communicating over a channel
hwswchan. The software establishes the network connection, and then
enters a loop which transmits and receives the position of the
local and remote plane, and sends new positions to the display
process.
[0201] Processor Design
[0202] The simple stack-based processor from Appendix 1 was
parameterized in the following ways to run this software. The width
of the processor was made to be 10 bits, which is sufficient to
address a character on the screen in a single word. No interrupts
were required, so these were removed, as were a number of unused
instructions, and the internal memory.
[0203] Co-simulation
[0204] The RT-level design was simulated using the Handel-C
simulator. Sample input files mimicking the expected inputs from
the peripherals were prepared, and these were fed into the
simulator. A black and white picture 700 of the color display is
shown in FIG. 7 (representing a snapshot of the X window drawn by
the co-simulator).
[0205] The design was then placed and routed using the proprietary
Xilinx tools, and successfully fit into the Xilinx 4013 FPGA on the
Hammond board.
[0206] This application would not have been easy to implement
without the codesign system of the invention. A hardware-only
solution would not have fitted. onto the FPGA; a software-only
solution would not have been able to generate the video and
interface with the ports at the required speed. The invention
allows the functionality of the target system to be partitioned
while parameterizing the processor to provide an optimal
system.
[0207] Real World Complications
[0208] The codesign system was presented with an implementation
challenge with this design. The processor had to access the RAM
(because that is where the program was stored), whilst the hardware
display process simultaneously had to access the RAM because this
is where the background picture, character map and screen map were
stored. This memory contention problem was made more difficult to
overcome because of an implementation decision made during the
design of the Hammond board: for a read cycle the synchronous
static RAM which was used requires the address to be presented the
cycle before the data is returned.
[0209] The display process needs to be able to access the memory
without delay, because of the tight timing constraints placed on
it. A semaphore is used to indicate when the display process
requires the memory. In this case the processor stalls until the
semaphore is lowered. On the next cycle the processor then presents
to the memory the address of the next instruction, which in some
cases may already have been presented once.
[0210] The designer was able to overcome this problem using the
codesign system of invention because of the facility for some
manual partitioning by the user and describing some parts of the
design at the register transfer level to give close control over
those parts. Thus while assisting the user, the system allows close
control where desired.
EXAMPLE 3
Mass-spring Simulation
[0211] Introduction
[0212] The "springs" program is a small example of a codesign
programmed in the C-like language mentioned above. It performs a
simulation of a simple mass-spring system, with a real time display
on a monitor, and interaction via a pair of buttons.
[0213] Design
[0214] The design consists of three parts: a process computing the
motion of the masses, a process rendering the positions of the
masses into line segments, and a process which displays these
segments and supplies the monitor with appropriate synchronization
signals. The first two processes are written in a single C-like
program. The display process is hard real-time and so requires a
language which can control external signals at the resolution of a
single clock cycle, so for this reason it is implemented using an
RTL description (Handel-C in this instance).
[0215] These two programs are shown in Appendix 3. They will be
explained below, together with the partitioning process and the
resulting implementation. FIG. 8 is a block diagram of the ultimate
implementation, together with a representation of the display of
the masses and springs. FIG. 9 is a dependency graph for
calculation of the variables required.
[0216] Mass Motion Process
[0217] The mass motion process first sets up the initial positions,
velocities and acceleration of the masses. This can be seen in
Appendix 3 where positions p0 to p7 are initialized as 65536. The
program then continues in an infinite loop, consisting of: sending
pairs of mass positions to the rendering process, computing updated
positions based on the velocities of the masses, computing updated
velocities based on the accelerations of the masses, and computing
accelerations based on the positions of the masses according to
Hooke's law. The process then reads the status of the control
buttons and sets the position of one of the masses accordingly.
This can be seen in Appendix 3 as the statement "received (buttons,
button status)".
[0218] This process is quite compute intensive over a short period
(requiring quite a number of operations to perform the motion
calculation), but since these only occur once per frame of video
the amortized time available for the calculation is quite long.
[0219] Rendering Process
[0220] The rendering process runs an infinite loop performing the
following operations: reading a pair of mass positions from the
mass motion process then interpolate in between these two positions
for the next 64 lines of video output. A pair of interpolated
positions is sent to the RTL display process once per line. This is
a relatively simple process with only one calculation, but this
must be performed very regularly.
[0221] Display Process
[0222] The display process (which is written in Handel-C) and is
illustrated in Appendix 3 reads start and end positions from the
rendering process and drives the video color signal between these
positions on a scan line. Simultaneously, it drives the
synchronization signals for the monitor. At the end of each frame
it reads the values from the external buttons and sends these to
the mass motion process.
[0223] Partitioning by the Codesign System
[0224] The design could be partitioned it in a large number of
ways. It could partition the entire design into hardware or into
software, partition the design at the high-level, by the first two
processes described above and compiling them using one of the
possible routes, or it can partition the design at a lower level,
and generate further parallel processes communicating with each
other. Whatever choice the partitioner makes, it maintains the
functional correctness of the design, but will change the cost of
the implementation (in terms of the area, clock cycles and so
forth). The user may direct the partitioner to choose one of the
options above the others. A number of the options are described
below.
[0225] Pure Hardware
[0226] The partitioner could map the first two processes directly
into Handel-C, after performing some additional parallelization.
The problem with this approach is that each one of the operations
in the mass motion process will be dedicated to its own piece of
hardware, in an effort to increase performance. However, as
discussed above, this is unnecessary as these calculations can be
performed at a slower speed. The result is a design that can
perform quickly enough but which is too large to fit on a single
FPGA. This problem would be recognized by the partitioner using its
area estimation techniques.
[0227] Pure Software
[0228] An alternative approach is for the partitioner to map the
two processes into software running on a parameterized threaded
processor. This reduces the area required, since the repeated
operations of the mass motion calculations are performed with a
single operation inside the processor. However, since the processor
must swap between doing the mass motion calculations and the
rendering calculations, overhead is introduced which causes it to
run too slowly to display in real-time. The partitioner can
recognize this by using the speed estimator, based on the profiling
information gathered from simulations of the system.
[0229] Software/Software
[0230] Another alternative would be for the partitioner to generate
a pair of parameterized processors running in parallel, the first
calculating motion and the second performing the rendering. The
area required is still smaller than the pure hardware approach, and
the speed is now sufficient to implement the system in real time.
However, using a parameterized processor for the rendering process
adds some overhead (for instance, performing the instruction
decoding), which is unnecessary. So although the solution works, it
is a sub optimal.
[0231] Hardware/Software
[0232] The best solution, and the one chosen by the partitioner, is
to partition the mass motion process into software for a
parameterized, unthreaded processor, and to partition the rendering
process 810 which was written at a behavioral level together with
the position, velocity and acceleration calculations 806 into
hardware. This solution has the minimum area of the options
considered, and performs sufficiently quickly to satisfy the real
time display process.
[0233] Thus referring to FIG. 8, the behavioral part of the system
802 includes the calculation of the positions, velocities and
accelerations of the masses at 806 (which will subsequently be
partitioned to software), and the line and drawing processes at 810
(which will subsequently be partitioned to hardware). The RTL
hardware 820 is used to receive the input from the buttons at 822
and output the video at 824.
[0234] Thus the partitioner 208 used the estimators 220, 222 and
224 to estimate the speed and area of each possible partition based
on the use of a customized processor. The interface cosynthesizer
210 implements the interface between hardware and software on two
FPGA channels 804 and 808 and these are used to transfer a position
information to the rendering process and to transfer the button
information to the position calculation 806 from button input
822.
[0235] The width adjuster 206, which is working on the mass motion
part of the problem to be partitioned to software, parameterizes
the processor to have a width of 17 bits and adjusts the width of
"curr_pos" which is the current position to nine bits, the width of
the segment channel. The processor parameterize at 17 further
parameterizes the processor by removing unused instructions such as
multiply, interrupts, and the data memory is reduced and
multi-threading is removed. Further, op codes are assigned and the
operator width is adjusted.
[0236] The description of the video output 824 and button interface
822 were, in this case, written in an RTL language, so there is no
behavioral synthesis to be done for them. Further, because the
hardware will be formed on an FPGA, no width adjustment is
necessary because the width can be set as desired.
[0237] The partitioner 208 generates a dependency graph as shown in
FIG. 9 which indicates which variables depend on which. It is used
by the partitioner to determine the communications costs associated
with the partitioning, for instance to assess the need for
variables to be passed from one resource to another given a
particular partitioning.
[0238] A preferred embodiment of a system in accordance with the
present invention is preferably practiced in the context of a
personal computer such as an IBM compatible personal computer,
Apple Macintosh computer or UNIX based workstation. A
representative hardware environment is depicted in FIG. 10, which
illustrates a typical hardware configuration of a workstation in
accordance with a preferred embodiment having a central processing
unit 1010, such as a microprocessor, and a number of other units
interconnected via a system bus 1012. The workstation shown in FIG.
10 includes a Random Access Memory (RAM) 1014, Read Only Memory
(ROM) 1016, an I/O adapter 1018 for connecting peripheral devices
such as disk storage units 1020 to the bus 1012, a user interface
adapter 1022 for connecting a keyboard 1024, a mouse 1026, a
speaker 1028, a microphone 1032, and/or other user interface
devices such as a touch screen (not shown) to the bus 1012,
communication adapter 1034 for connecting the workstation to a
communication network (e.g., a data processing network) and a
display adapter 1036 for connecting the bus 1012 to a display
device 1038. The workstation typically has resident thereon an
operating system such as the Microsoft Windows NT or Windows/95
Operating System (OS), the IBM OS/2 operating system, the MAC OS,
or UNIX operating system. Those skilled in the art will appreciate
that the present invention may also be implemented on platforms and
operating systems other than those mentioned.
[0239] A preferred embodiment is written using JAVA, C, and the C++
language and utilizes object oriented programming methodology.
Object oriented programming (OOP) has become increasingly used to
develop complex applications. As OOP moves toward the mainstream of
software design and development, various software solutions require
adaptation to make use of the benefits of OOP. A need exists for
these principles of OOP to be applied to a messaging interface of
an electronic messaging system such that a set of OOP classes and
objects for the messaging interface can be provided.
[0240] OOP is a process of developing computer software using
objects, including the steps of analyzing the problem, designing
the system, and constructing the program. An object is a software
package that contains both data and a collection of related
structures and procedures. Since it contains both data and a
collection of structures and procedures, it can be visualized as a
self-sufficient component that does not require other additional
structures, procedures or data to perform its specific task. OOP,
therefore, views a computer program as a collection of largely
autonomous components, called objects, each of which is responsible
for a specific task. This concept of packaging data, structures,
and procedures together in one component or module is called
encapsulation.
[0241] In general, OOP components are reusable software modules
which present an interface that conforms to an object model and
which are accessed at run-time through a component integration
architecture. A component integration architecture is a set of
architecture mechanisms which allow software modules in different
process spaces to utilize each others capabilities or functions.
This is generally done by assuming a common component object model
on which to build the architecture. It is worthwhile to
differentiate between an object and a class of objects at this
point. An object is a single instance of the class of objects,
which is often just called a class. A class of objects can be
viewed as a blueprint, from which many objects can be formed.
[0242] OOP allows the programmer to create an object that is a part
of another object. For example, the object representing a piston
engine is said to have a composition-relationship with the object
representing a piston. In reality, a piston engine comprises a
piston, valves and many other components; the fact that a piston is
an element of a piston engine can be logically and semantically
represented in OOP by two objects.
[0243] OOP also allows creation of an object that "depends from"
another object. If there are two objects, one representing a piston
engine and the other representing a piston engine wherein the
piston is made of ceramic, then the relationship between the two
objects is not that of composition. A ceramic piston engine does
not make up a piston engine. Rather it is merely one kind of piston
engine that has one more limitation than the piston engine; its
piston is made of ceramic. In this case, the object representing
the ceramic piston engine is called a derived object, and it
inherits all of the aspects of the object representing the piston
engine and adds further limitation or detail to it. The object
representing the ceramic piston engine "depends from" the object
representing the piston engine. The relationship between these
objects is called inheritance.
[0244] When the object or class representing the ceramic piston
engine inherits all of the aspects of the objects representing the
piston engine, it inherits the thermal characteristics of a
standard piston defined in the piston engine class. However, the
ceramic piston engine object overrides these ceramic specific
thermal characteristics, which are typically different from those
associated with a metal piston. It skips over the original and uses
new functions related to ceramic pistons. Different kinds of piston
engines have different characteristics, but may have the same
underlying functions associated with it (e.g., how many pistons in
the engine, ignition sequences, lubrication, etc.). To access each
of these functions in any piston engine object, a programmer would
call the same functions with the same names, but each type of
piston engine may have different/overriding implementations of
functions behind the same name. This ability to hide different
implementations of a function behind the same name is called
polymorphism and it greatly simplifies communication among
objects.
[0245] With the concepts of composition-relationship,
encapsulation, inheritance and polymorphism, an object can
represent just about anything in the real world. In fact, one's
logical perception of the reality is the only limit on determining
the kinds of things that can become objects in object-oriented
software. Some typical categories are as follows:
[0246] Objects can represent physical objects, such as automobiles
in a traffic-flow simulation, electrical components in a
circuit-design program, countries in an economics model, or
aircraft in an air-traffic-control system.
[0247] Objects can represent elements of the computer-user
environment such as windows, menus or graphics objects.
[0248] An object can represent an inventory, such as a personnel
file or a table of the latitudes and longitudes of cities.
[0249] An object can represent user-defined data types such as
time, angles, and complex numbers, or points on the plane.
[0250] With this enormous capability of an object to represent just
about any logically separable matters, OOP allows the software
developer to design and implement a computer program that is a
model of some aspects of reality, whether that reality is a
physical entity, a process, a system, or a composition of matter.
Since the object can represent anything, the software developer can
create an object which can be used as a component in a larger
software project in the future.
[0251] If 90% of a new OOP software program consists of proven,
existing components made from preexisting reusable objects, then
only the remaining 10% of the new software project has to be
written and tested from scratch. Since 90% already came from an
inventory of extensively tested reusable objects, the potential
domain from which an error could originate is 10% of the program.
As a result, OOP enables software developers to build objects out
of other, previously built objects.
[0252] This process closely resembles complex machinery being built
out of assemblies and sub-assemblies. OOP technology, therefore,
makes software engineering more like hardware engineering in that
software is built from existing components, which are available to
the developer as objects. All this adds up to an improved quality
of the software as well as an increased speed of its
development.
[0253] Programming languages are beginning to fully support the OOP
principles, such as encapsulation, inheritance, polymorphism, and
composition-relationship. With the advent of the C++ language, many
commercial software developers have embraced OOP. C++ is an OOP
language that offers a fast, machine-executable code. Furthermore,
C++ is suitable for both commercial-application and
systems-programming projects. For now, C++ appears to be the most
popular choice among many OOP programmers, but there is a host of
other OOP languages, such as Smalltalk, Common Lisp Object System
(CLOS), and Eiffel. Additionally, OOP capabilities are being added
to more traditional popular computer programming languages such as
Pascal.
[0254] The benefits of object classes can be summarized, as
follows:
[0255] Objects and their corresponding classes break down complex
programming problems into many smaller, simpler problems.
[0256] Encapsulation enforces data abstraction through the
organization of data into small, independent objects that can
communicate with each other. Encapsulation protects the data in an
object from accidental damage, but allows other objects to interact
with that data by calling the object's member functions and
structures.
[0257] Subclassing and inheritance make it possible to extend and
modify objects through deriving new kinds of objects from the
standard classes available in the system. Thus, new capabilities
are created without having to start from scratch.
[0258] Polymorphism and multiple inheritance make it possible for
different programmers to mix and match characteristics of many
different classes and create specialized objects that can still
work with related objects in predictable ways.
[0259] Class hierarchies and containment hierarchies provide a
flexible mechanism for modeling real-world objects and the
relationships among them.
[0260] Libraries of reusable classes are useful in many situations,
but they also have some limitations. For example:
[0261] Complexity. In a complex system, the class hierarchies for
related classes can become extremely confusing, with many dozens or
even hundreds of classes.
[0262] Flow of control. A program written with the aid of class
libraries is still responsible for the flow of control (i.e., it
must control the interactions among all the objects created from a
particular library). The programmer has to decide which functions
to call at what times for which kinds of objects.
[0263] Duplication of effort. Although class libraries allow
programmers to use and reuse many small pieces of code, each
programmer puts those pieces together in a different way. Two
different programmers can use the same set of class libraries to
write two programs that do exactly the same thing but whose
internal structure (i.e., design) may be quite different, depending
on hundreds of small decisions each programmer makes along the way.
Inevitably, similar pieces of code end up doing similar things in
slightly different ways and do not work as well together as they
should.
[0264] Class libraries are very flexible. As programs grow more
complex, more programmers are forced to reinvent basic solutions to
basic problems over and over again. A relatively new extension of
the class library concept is to have a framework of class
libraries. This framework is more complex and consists of
significant collections of collaborating classes that capture both
the small scale patterns and major mechanisms that implement the
common requirements and design in a specific application domain.
They were first developed to free application programmers from the
chores involved in displaying menus, windows, dialog boxes, and
other standard user interface elements for personal computers.
[0265] Frameworks also represent a change in the way programmers
think about the interaction between the code they write and code
written by others. In the early days of procedural programming, the
programmer called libraries provided by the operating system to
perform certain tasks, but basically the program executed down the
page from start to finish, and the programmer was solely
responsible for the flow of control. This was appropriate for
printing out paychecks, calculating a mathematical table, or
solving other problems with a program that executed in just one
way.
[0266] The development of graphical user interfaces began to turn
this procedural programming arrangement inside out. These
interfaces allow the user, rather than program logic, to drive the
program and decide when certain actions should be performed. Today,
most personal computer software accomplishes this by means of an
event loop which monitors the mouse, keyboard, and other sources of
external events and calls the appropriate parts of the programmer's
code according to actions that the user performs. The programmer no
longer determines the order in which events occur. Instead, a
program is divided into separate pieces that are called at
unpredictable times and in an unpredictable order. By relinquishing
control in this way to users, the developer creates a program that
is much easier to use. Nevertheless, individual pieces of the
program written by the developer still call libraries provided by
the operating system to accomplish certain tasks, and the
programmer must still determine the flow of control within each
piece after it's called by the event loop. Application code still
"sits on top of" the system.
[0267] Even event loop programs require programmers to write a lot
of code that should not need to be written separately for every
application. The concept of an application framework carries the
event loop concept further. Instead of dealing with all the nuts
and bolts of constructing basic menus, windows, and dialog boxes
and then making these things all work together, programmers using
application frameworks start with working application code and
basic user interface elements in place. Subsequently, they build
from there by replacing some of the generic capabilities of the
framework with the specific capabilities of the intended
application.
[0268] Application frameworks reduce the total amount of code that
a programmer has to write from scratch. However, because the
framework is really a generic application that displays windows,
supports copy and paste, and so on, the programmer can also
relinquish control to a greater degree than event loop programs
permit. The framework code takes care of almost all event handling
and flow of control, and the programmer's code is called only when
the framework needs it (e.g., to create or manipulate a proprietary
data structure).
[0269] A programmer writing a framework program not only
relinquishes control to the user (as is also true for event loop
programs), but also relinquishes the detailed flow of control
within the program to the framework. This approach allows the
creation of more complex systems that work together in interesting
ways, as opposed to isolated programs, having custom code, being
created over and over again for similar problems.
[0270] Thus, as is explained above, a framework basically is a
collection of cooperating classes that make up a reusable design
solution for a given problem domain. It typically includes objects
that provide default behavior (e.g., for menus and windows), and
programmers use it by inheriting some of that default behavior and
overriding other behavior so that the framework calls application
code at the appropriate times.
[0271] There are three main differences between frameworks and
class libraries:
[0272] Behavior versus protocol. Class libraries are essentially
collections of behaviors that you can call when you want those
individual behaviors in your program. A framework, on the other
hand, provides not only behavior but also the protocol or set of
rules that govern the ways in which behaviors can be combined,
including rules for what a programmer is supposed to provide versus
what the framework provides.
[0273] Call versus override. With a class library, the code the
programmer instantiates objects and calls their member functions.
It's possible to instantiate and call objects in the same way with
a framework (i.e., to treat the framework as a class library), but
to take full advantage of a framework's reusable design, a
programmer typically writes code that overrides and is called by
the framework. The framework manages the flow of control among its
objects. Writing a program involves dividing responsibilities among
the various pieces of software that are called by the framework
rather than specifying how the different pieces should work
together.
[0274] Implementation versus design. With class libraries,
programmers reuse only implementations, whereas with frameworks,
they reuse design. A framework embodies the way a family of related
programs or pieces of software work. It represents a generic design
solution that can be adapted to a variety of specific problems in a
given domain. For example, a single framework can embody the way a
user interface works, even though two different user interfaces
created with the same framework might solve quite different
interface problems.
[0275] Thus, through the development of frameworks for solutions to
various problems and programming tasks, significant reductions in
the design and development effort for software can be achieved. A
preferred embodiment of the invention utilizes HyperText Markup
Language (HTML) to implement documents on the Internet together
with a general-purpose secure communication protocol for a
transport medium between the client and the Newco. HTTP or other
protocols could be readily substituted for HTML without undue
experimentation. Information on these products is available in T.
Berners-Lee, D. Connoly, "RFC 1866: Hypertext Markup Language--2.0"
(Nov. 1995); and R. Fielding, H, Frystyk, T. Berners-Lee, J. Gettys
and J.C. Mogul, "Hypertext Transfer Protocol--HTTP/1.1: HTTP
Working Group Internet Draft" (May 2, 1996). HTML is a simple data
format used to create hypertext documents that are portable from
one platform to another. HTML documents are SGML documents with
generic semantics that are appropriate for representing information
from a wide range of domains. HTML has been in use by the
World-Wide Web global information initiative since 1990. HTML is an
application of ISO Standard 8879; 1986 information Processing Text
and Office Systems; Standard Generalized Markup Language
(SGML).
[0276] To date, Web development tools have been limited in their
ability to create dynamic Web applications which span from client
to server and interoperate with existing computing resources. Until
recently, HTML has been the dominant technology used in development
of Web-based solutions. However, HTML has proven to be inadequate
in the following areas:
[0277] Poor performance;
[0278] Restricted user interface capabilities;
[0279] Can only produce static Web pages;
[0280] Lack of interoperability with existing applications and
data; and
[0281] Inability to scale.
[0282] Sun Microsystem's Java language solves many of the
client-side problems by:
[0283] Improving performance on the client side;
[0284] Enabling the creation of dynamic, real-time Web
applications; and
[0285] Providing the ability to create a wide variety of user
interface components.
[0286] With Java, developers can create robust User Interface (UI)
components. Custom "widgets" (e.g., real-time stock tickers,
animated icons, etc.) can be created, and client-side performance
is improved. Unlike HTML, Java supports the notion of client-side
validation, offloading appropriate processing onto the client for
improved performance. Dynamic, real-time Web pages can be created.
Using the above-mentioned custom Ul components, dynamic Web pages
can also be created.
[0287] Sun's Java language has emerged as an industry-recognized
language for "programming the Internet." Sun defines Java as: "a
simple, object-oriented, distributed, interpreted, robust, secure,
architecture-neutral, portable, high-performance, multithreaded,
dynamic, buzzword-compliant, general-purpose programming language.
Java supports programming for the Internet in the form of
platform-independent Java applets." Java applets are small,
specialized applications that comply with Sun's Java Application
Programming Interface (API) allowing developers to add "interactive
content" to Web documents (e.g., simple animations, page
adornments, basic games, etc.). Applets execute within a
Java-compatible browser (e.g., Netscape Navigator) by copying code
from the server to client. From a language standpoint, Java's core
feature set is based on C++. Sun's Java literature states that Java
is basically, "C++ with extensions from Objective C for more
dynamic method resolution."
[0288] Another technology that provides similar function to JAVA is
provided by Microsoft and ActiveX Technologies, to give developers
and Web designers wherewithal to build dynamic content for the
Internet and personal computers. ActiveX includes tools for
developing animation, 3-D virtual reality, video and other
multimedia content. The tools use Internet standards, work on
multiple platforms, and are being supported by over 100 companies.
The group's building blocks are called ActiveX Controls, small,
fast components that enable developers to embed parts of software
in hypertext markup language (HTML) pages. ActiveX Controls work
with a variety of programming languages including Microsoft Visual
C++, Borland Delphi, Microsoft Visual Basic programming system and,
in the future, Microsoft's development tool for Java, code named
"Jakarta." ActiveX Technologies also includes ActiveX Server
Framework, allowing developers to create server applications. One
of ordinary skill in the art readily recognizes that ActiveX could
be substituted for JAVA without undue experimentation to practice
the invention.
[0289] Summary
[0290] Thus the codesign system of the invention has the following
advantages in designing a target system:
[0291] 1. It uses parameterization and instruction addition and
removal for optimal processor design in on FPGA. The system
provides an environment in which an FPGA-based processor and its
compiler can be developed in a single framework.
[0292] 2. It can generate designs containing multiple communicating
processors. parameterized custom processors, and the
inter-processor communication can be tuned for the application.
[0293] 3. The hardware can be designed to run in parallel with the
processors to meet speed constraints. Thus time critical parts of
the system can be allocated to custom hardware, which can be
designed at the behavioral or register transfer level.
[0294] 4. Non-time critical parts of the design can be allocated to
software, and run on a small, slow processor.
[0295] 5. The system can target circuitry on dynamic FPGAs. The
FPGA can contain a small processor which can configure and
reconfigure the rest of the FPGA at run time.
[0296] 6. The system allows the user to explore efficient system
implementations, by allowing parameterized application-specific
processors with user-defined instructions to communicate with
custom hardware. This combination of custom processor and custom
hardware allows a very large design space to be explored by the
user.
[0297] C Functions in Hardware
[0298] FIG. 11 depicts a process 1100 for compiling a C function to
a reconfigurable logic device. In operation 1102, a function
written in a C programming language is received. The C function is
compiled into processor instructions in operation 1104. In
operation 1106, the processor instructions are used to generate
hardware configuration information. In operation 1108, a Field
Programmable Gate Array (FPGA) is configured using the hardware
configuration information such that the function is compiled to the
FPGA. Note that the methodology of the present invention could also
be applied to compile functions to reconfigurable logic devices
other than FPGAs.
[0299] A system for compiling a C function to a reconfigurable
logic device is also provided. The system includes receiving logic
for receiving a function written in a C programming language.
Compiling logic is used to compile the C function into processor
instructions. Conversion logic generates hardware configuration
information from the processor instructions. Configuring logic
utilizes the hardware configuration information to configure an
FPGA such that the function is compiled to the FPGA.
[0300] In one embodiment of the present invention, the function is
a shared function. More particularly, the function in the FPGA is
shared amongst all its uses. In another embodiment of the present
invention, the configuration of the FPGA is duplicated for each
use, so that the function is used as an inline function. In yet
another embodiment of the present invention, the FPGA is configured
to provide an array of functions, where N copies of the function
are specified for use M times.
[0301] In a preferred embodiment of the present invention, a token
is used to invoke the function. Preferably, when invoking the
function, the token is passed to a start signal, the start signal
and call data are routed to the function, and the token is stored
in a wait sub-circuit until the function is completed.
[0302] Handel-C is the preferred programming language for carrying
out the methodology of the present invention and configuring the
FPGA. One skilled in the art will be familiar with programming in
Handel-C and therefore only a general discussion of Handel-C will
be provided. Handel-C is described in more detail below in the
section entitled "Handel-C."
[0303] Three illustrative types of functions are declared as
follows in Handel-C:
6 Shared function void f(void); Inline function inline void
f(void); Array of functions void f[n](void);
[0304] These functions are invoked as follows:
7 f( ); causes logic to be built calling the only function
implementation. inline f( ); causes a new circuit implementing the
function to be built. f[3]( ); causes logic to be built calling the
third implementation of the function.
[0305] FIG. 12 illustrates the control logic 1200 for calling
(invoking) functions which are shared. Handel-C circuits are
generally controlled by tokens. A function is called by passing a
token to the START signal 1202. The multiplexer 1204 routes the
START signal and associated data from this call to the
implementation of the function body 1206. The token is stored in a
"wait sub-circuit" 1208. The wait sub-circuit includes an OR gate
1212, an AND gate 1214 with an inverter, a second AND gate 1216,
and a flip-flop 1218 which stores the token. When the function is
completed, the DONE signal 1210 is asserted and the token is passed
to the circuitry following this invocation of the function.
[0306] FIG. 13 depicts a pass by value sub-circuit 1300 according
to an embodiment of the present invention. Passing by value uses
the circuit which, on the first clock cycle of the function body
1206 (see FIG. 12), copies the values of the arguments into
temporary variables, unless those parameters are written to in the
first clock cycle, in which case the value being written is stored
in the variable in that cycle. With continued reference to FIG. 13,
the temporary variable is stored in a storage medium 1302, which
can include memory or reconfigurable logic for example. Gating
logic 1304 handles the write to the temporary variable. Note that
the gating logic includes several AND and OR gates. A multiplexer
1306 handles the read from the parameter, wich is either a
temporary variable or the value passed in. FIRST is a signal that
is true if it is the first clock cycle of the function call.
D.sub.A is the argument to the function. D.sub.1 and D.sub.2 are
the other data written to the variable, together with the
associated write enables (WE).
[0307] Reconfigurable Logic Devices
[0308] Field-Programmable Logic Devices (FPLD's) have continuously
evolved to better serve the unique needs of different end-users.
From the time of introduction of simple PLD's such as the Advanced
Micro Devices 22V10.TM. Programmable Array Logic device (PAL), the
art has branched out in several different directions and
bloomed.
[0309] One evolutionary branch of FPLD's has grown along a paradigm
known as Complex PLD's or CPLD's. This paradigm is characterized by
devices such as the Advanced Micro Devices MACH.TM. family.
Examples of CPLD circuitry are seen in U.S. Pat. No. 5,015,884
(issued May 14, 1991 to Om P. Agrawal et al.) and U.S. Pat. No.
5,151,623 (issued Sep. 29, 1992 to Om P. Agrawal et al.), which are
herein incorporated by reference.
[0310] Another evolutionary chain in the art of field programmable
logic has branched out along a paradigm known as Field Programmable
Gate Arrays or FPGA's. Examples of such devices include the
XC2000.TM. and XC3000.TM. families of FPGA devices introduced by
Xilinx, Inc. of San Jose, Calif. The architectures of these devices
are exemplified in U.S. Pat. Nos. 4,642,487; 4,706,216; 4,713,557;
and 4,758,985; each of which is originally assigned to Xilinx, Inc.
and which are herein incorporated by reference for all
purposes.
[0311] An FPGA device can be characterized as an integrated circuit
that has four major features as follows.
[0312] (1) A user-accessible, configuration-defining memory means,
such as SRAM, PROM, EPROM, EEPROM, anti-fused, fused, or other, is
provided in the FPGA device so as to be at least once-programmable
by device users for defining user-provided configuration
instructions. Static Random Access Memory or SRAM is of course, a
form of reprogrammable memory that can be differently programmed
many times. Electrically Erasable and reProgrammable ROM or EEPROM
is an example of nonvolatile reprogrammable memory. The
configuration-defining memory of an FPGA device can be formed of
mixture of different kinds of memory elements if desired (e.g.,
SRAM and EEPROM) although this is not a popular approach.
[0313] (2) Input/Output Blocks (IOB's) are provided for
interconnecting other internal circuit components of the FPGA
device with external circuitry. The IOB's' may have fixed
configurations or they may be configurable in accordance with
user-provided configuration instructions stored in the
configuration-defining memory means.
[0314] (3) Configurable Logic Blocks (CLB's) are provided for
carrying out user-programmed logic functions as defined by
user-provided configuration instructions stored in the
configuration-defining memory means.
[0315] Typically, each of the many CLB's of an FPGA has at least
one lookup table (LUT) that is user-configurable to define any
desired truth table,--to the extent allowed by the address space of
the LUT. Each CLB may have other resources such as LUT input signal
pre-processing resources and LUT output signal post-processing
resources. Although the term `CLB` was adopted by early pioneers of
FPGA technology, it is not uncommon to see other names being given
to the repeated portion of the FPGA that carries out
user-programmed logic functions. The term, `LAB` is used for
example in U.S. Pat. No. 5,260,611 to refer to a repeated unit
having a 4-input LUT.
[0316] (4) An interconnect network is provided for carrying signal
traffic within the FPGA device between various CLB's and/or between
various IOB's and/or between various IOB's and CLB's. At least part
of the interconnect network is typically configurable so as to
allow for programmably-defined routing of signals between various
CLB's and/or IOB's in accordance with user-defined routing
instructions stored in the configuration-defining memory means.
[0317] In some instances, FPGA devices may additionally include
embedded volatile memory for serving as scratchpad memory for the
CLB's or as FIFO or LIFO circuitry. The embedded volatile memory
may be fairly sizable and can have 1 million or more storage bits
in addition to the storage bits of the device's configuration
memory.
[0318] Modern FPGA's tend to be fairly complex. They typically
offer a large spectrum of user-configurable options with respect to
how each of many CLB's should be configured, how each of many
interconnect resources should be configured, and/or how each of
many IOB's should be configured. This means that there can be
thousands or millions of configurable bits that may need to be
individually set or cleared during configuration of each FPGA
device.
[0319] Rather than determining with pencil and paper how each of
the configurable resources of an FPGA device should be programmed,
it is common practice to employ a computer and appropriate
FPGA-configuring software to automatically generate the
configuration instruction signals that will be supplied to, and
that will ultimately cause an unprogrammed FPGA to implement a
specific design. (The configuration instruction signals may also
define an initial state for the implemented design, that is,
initial set and reset states for embedded flip flops and/or
embedded scratchpad memory cells.)
[0320] The number of logic bits that are used for defining the
configuration instructions of a given FPGA device tends to be
fairly large (e.g., 1 Megabits or more) and usually grows with the
size and complexity of the target FPGA. Time spent in loading
configuration instructions and verifying that the instructions have
been correctly loaded can become significant, particularly when
such loading is carried out in the field.
[0321] For many reasons, it is often desirable to have in-system
reprogramming capabilities so that reconfiguration of FPGA's can be
carried out in the field.
[0322] FPGA devices that have configuration memories of the
reprogrammable kind are, at least in theory, `in-system
programmable` (ISP). This means no more than that a possibility
exists for changing the configuration instructions within the FPGA
device while the FPGA device is `in-system` because the
configuration memory is inherently reprogrammable. The term,
`in-system` as used herein indicates that the FPGA device remains
connected to an application-specific printed circuit board or to
another form of end-use system during reprogramming. The end-use
system is of course, one which contains the FPGA device and for
which the FPGA device is to be at least once configured to operate
within in accordance with predefined, end-use or `in the field`
application specifications.
[0323] The possibility of reconfiguring such inherently
reprogrammable FPGA's does not mean that configuration changes can
always be made with any end-use system. Nor does it mean that,
where in-system reprogramming is possible, that reconfiguration of
the FPGA can be made in timely fashion or convenient fashion from
the perspective of the end-use system or its users. (Users of the
end-use system can be located either locally or remotely relative
to the end-use system.)
[0324] Although there may be many instances in which it is
desirable to alter a pre-existing configuration of an in the field
FPGA (with the alteration commands coming either from a remote site
or from the local site of the FPGA), there are certain practical
considerations that may make such in-system reprogrammability of
FPGA's more difficult than first apparent (that is, when
conventional techniques for FPGA reconfiguration are followed).
[0325] A popular class of FPGA integrated circuits (IC's) relies on
volatile memory technologies such as SRAM (static random access
memory) for implementing on-chip configuration memory cells. The
popularity of such volatile memory technologies is owed primarily
to the inherent reprogrammability of the memory over a device
lifetime that can include an essentially unlimited number of
reprogramming cycles.
[0326] There is a price to be paid for these advantageous features,
however. The price is the inherent volatility of the configuration
data as stored in the FPGA device. Each time power to the FPGA
device is shut off, the volatile configuration memory cells lose
their configuration data. Other events may also cause corruption or
loss of data from volatile memory cells within the FPGA device.
[0327] Some form of configuration restoration means is needed to
restore the lost data when power is shut off and then re-applied to
the FPGA or when another like event calls for configuration
restoration (e.g., corruption of state data within scratchpad
memory).
[0328] The configuration restoration means can take many forms. If
the FPGA device resides in a relatively large system that has a
magnetic or optical or opto-magnetic form of nonvolatile memory
(e.g., a hard magnetic disk)--and the latency of powering up such a
optical/magnetic device and/or of loading configuration
instructions from such an optical/magnetic form of nonvolatile
memory can be tolerated--then the optical/magnetic memory device
can be used as a nonvolatile configuration restoration means that
redundantly stores the configuration data and is used to reload the
same into the system's FPGA device(s) during power-up operations
(and/or other restoration cycles).
[0329] On the other hand, if the FPGA device(s) resides in a
relatively small system that does not have such optical/magnetic
devices, and/or if the latency of loading configuration memory data
from such an optical/magnetic device is not tolerable, then a
smaller and/or faster configuration restoration means may be called
for.
[0330] Many end-use systems such as cable-TV set tops, satellite
receiver boxes, and communications switching boxes are constrained
by prespecified design limitations on physical size and/or power-up
timing and/or security provisions and/or other provisions such that
they cannot rely on magnetic or optical technologies (or on
network/satellite downloads) for performing configuration
restoration. Their designs instead call for a relatively small and
fast acting, non-volatile memory device (such as a
securely-packaged EPROM IC), for performing the configuration
restoration function. The small/fast device is expected to satisfy
application-specific criteria such as: (1) being securely retained
within the end-use system; (2) being able to store FPGA
configuration data during prolonged power outage periods; and (3)
being able to quickly and automatically re-load the configuration
instructions back into the volatile configuration memory (SRAM) of
the FPGA device each time power is turned back on or another event
calls for configuration restoration.
[0331] The term `CROP device` will be used herein to refer in a
general way to this form of compact, nonvolatile, and fast-acting
device that performs `Configuration-Restoring On Power-up` services
for an associated FPGA device.
[0332] Unlike its supported, volatilely reprogrammable FPGA device,
the corresponding CROP device is not volatile, and it is generally
not `in-system programmable`. Instead, the CROP device is generally
of a completely nonprogrammable type such as exemplified by
mask-programmed ROM IC's or by once-only programmable, fuse-based
PROM IC's. Examples of such CROP devices include a product family
that the Xilinx company provides under the designation `Serial
Configuration PROMs` and under the trade name, XC1700D.TM. These
serial CROP devices employ one-time programmable PROM (Programmable
Read Only Memory) cells for storing configuration instructions in
nonvolatile fashion.
[0333] Handel-C
[0334] C is a widely used programming language described in "The C
Programming Language", Brian Kernighan and Dennis Ritchie, Prentice
Hall 1988. Standard techniques exist for the compilation of C into
processor instructions such as "Compilers: Principles, Techniques
and Tools", Aho, Sethi and Ullman, Addison Wesley 1998, and
"Advanced Compiler Design and Implementation", Steven Muchnik,
Morgan Kauffman 1997, which are herein incorporated by
reference.
[0335] Handel was a programming language designed for compilation
into custom synchronous hardware, which was first described in
"Compiling occam into FPGAs", Ian Page and Wayne Luk in "FPGAs"
Eds. Will Moore and Wayne Luk, pp 271-283, Abingdon EE & CS
Books, 1991, which are herein incorporated by reference. Handel was
later given a C-like syntax (described in "Advanced Silicon
Prototyping in a Reconfigurable Environment", M. Aubury, I. Page,
D. Plunkett, M. Sauer and J. Saul, Proceedings of WoTUG 98, 1998,
which is also incorporated by reference), to produce several
versions of Handel-C.
[0336] Handel-C is the preferred programming language for carrying
out the methodology of the present invention and configuring the
FPGA. Handel-C is a programming language marketed by Celoxica
Limited, 7-8 Milton Park, Abingdon, Oxfordshire, OX14 4RT, United
Kingdom. It enables a software or hardware engineer to target
directly FPGAs in a similar fashion to classical microprocessor
cross-compiler development tools, without recourse to a Hardware
Description Language, thereby allowing the designer to directly
realize the raw real-time computing capability of the FPGA.
[0337] Handel-C is designed to enable the compilation of programs
into synchronous hardware; it is aimed at compiling high level
algorithms directly into gate level hardware.
[0338] The Handel-C syntax is based on that of conventional C so
programmers familiar with conventional C will recognize almost all
the constructs in the Handel-C language.
[0339] Sequential programs can be written in Handel-C just as in
conventional C but to gain the most benefit in performance from the
target hardware its inherent parallelism must be exploited.
[0340] Handel-C includes parallel constructs that provide the means
for the programmer to exploit this benefit in his applications. The
compiler compiles and optimizes Handel-C source code into a file
suitable for simulation or a netlist which can be placed and routed
on a real FPGA.
[0341] The simulator allows a user to test a program without using
real hardware. It can display the state of every variable
(register) in your program at every clock cycle if required, the
simulation steps and the number of cycles simulated being under
program control. Optionally the source code that was executed at
each clock cycle as well as the program state may be displayed in
order to assist in the debugging of the source code.
[0342] Further debugging options are provided in the toolset,
notably the `Logic Estimator`. This tool displays the source code
in a color highlighted form which relates to the logic depth and
usage. So providing feedback to the designer for further
optimizations.
[0343] While various embodiments have been described above, it
should be understood that they have been presented by way of
example only, and not limitation. Thus, the breadth and scope of a
preferred embodiment should not be limited by any of the above
described exemplary embodiments, but should be defined only in
accordance with the following claims and their equivalents.
* * * * *